Design
The initial design phase focused on exploring the potential protein-protein interaction between the bacterial enzyme Imidazole glycerol phosphate synthase (hisF) and the human AMP-activated protein kinase (AMPK).
The central hypothesis was that hisF from Lactobacillus rhamnosus inhibit adipogenesis by inducing AMPK in wetlab result than any other species that validating the result by investigating structural analysis of several other species.
To investigate this, several orthologs of the hisF protein were selected from different bacterial species for computational analysis:- Lactobacillus rhamnosus GG
- Staphylococcus epidermis
- Escherichia coli K-12
- Staphylococcus simulans
The primary design inputs for the subsequent build phase were the specific amino acid sequences for each of these hisF proteins and for AMPK.
Build
The “build” phase was conducted entirely in silico, translating the designed protein sequences into predicted three-dimensional structural models. Two state-of-the-art computational tools were used to generate the protein complex structures:
1. ColabFold (AlphaFold2): The sequences of each hisF variant and AMPK were input into ColabFold. The alphafold2_multimer_v3 model was specifically selected to predict the structure of the hetero-oligomeric complex. The process was executed by running the prediction pipeline for each pair.
2. AlphaFold Server (AlphaFold3): To corroborate and compare the initial findings, the same protein sequence pairs were submitted to the more recent AlphaFold3 server. This platform allows for the separate input of multiple protein entities to model their interactions.
Figure 1. Access ColabFold and paste the sequence (Link: www.colabfold.com)
Figure 2. Select the MSA options, model type, and other settings
Figrue 3. Push “Run all”
Figrue 4. Alphafold3: access to AlphaFold server and paste sequences (Link: www.alphafoldserver.com)
Figrue 5. Alphafold3: browse through the results
Test
The computationally generated models were then rigorously tested and evaluated based on the confidence metrics produced by the AlphaFold algorithms. The goal was to determine the reliability of the predicted structures, with a specific focus on the interface between hisF and AMPK.
- Interface Predicted Template Modeling (ipTM) Score: This metric was the primary indicator of confidence in the predicted interaction.
- Predicted Local Distance Difference Test (pLDDT): This score measures the confidence in the local structure of individual protein chains on a per-residue basis. Models were color-coded, with blue indicating very high confidence (>90) and red indicating very low confidence (<50). While parts of the individual proteins folded with high confidence, this metric does not guarantee a confident prediction of the interaction interface.
- Predicted Aligned Error (PAE): PAE plots were analyzed to assess the model’s confidence in the relative positions and orientations of the different protein domains. High error regions between the two protein chains further suggested a lack of a well-defined interaction.
Figrue 6. hisF of Lactobacillus Rhamnosus GG and human AMPK
Top right figure (PAE plot): PAE shows predicted error (in Å) in the relative position of residue i when the model is aligned on residue j
middle predicted lDDT plot: per-residue confidence of prediction on a 0 – 100 scale. Tells whether each region of the sequence is predicted well
- > 90 : very high confidence
- 70–90 : generally reliable
- 50–70 : low confidence / possibly flexible
- < 50 : likely disordered / unreliable
Bottom right figure (coverage plot): shows the multiple sequence alignment (MSA) used for prediction. - First row: sequences aligned to the both hisF and AMPK - Second & Third row: sequences aligned to either hisF or AMPK exclusively
iPTM: interfacial predicted TM-score (a metric AlphaFold-Multimer uses to quantify confidence in the predicted interface between chains)
Figrue 7. hisF of Staphylococcus epidermis and human AMPK
Figrue 8. hisF of Escherichia coli k12 and human AMPK
Figrue 9. hisF of Staphylococcus simulans and human AMPK
Figrue 10. hisF of Lactobacillus Rhamnosus GG and human AMPK (AlphaFold3)
Figrue 11. hisF of Staphylococcus epidermis and human AMPK (AlphaFold3)
Figrue 12. hisF of Escherichia coli k12 and human AMPK (AlphaFold3)
Figrue 13. hisF of Staphylococcus simulans and human AMPK (AlphaFold3)
Figrue 14. Superposed hisFs of Bacteria collection
Learn
The analysis of the test metrics provided a clear conclusion: both AlphaFold2 and AlphaFold3 predict with low confidence that the selected hisF proteins form a stable complex with AMPK. The consistently low ipTM scores strongly suggest that a direct, stable binding event is unlikely, however, still, ipTM of Lactobacillus rhamnosus has the highest score than other species, which validate our wetlab experiments that hisF from Lactobacillus rhamnosus is different from other strains in structural integration level.
This outcome is a critical learning step. It indicates that proceeding to wet lab experiments (e.g., protein expression, purification, and binding assays like pull-downs or calorimetry) with these specific protein pairs would have a low probability of success. This computational cycle effectively saved time and resources. The knowledge gained now informs the next “Design” phase, whether hisF interact with AMPK via another pathway